Sensor Based Real-Time Scheduling in Thermally Constrained Uniprocessor Systems

Due to increasing power densities, thermal management is becoming an important issue in real-time systems. In particular, without thermal-aware scheduling, execution of tasks in a real-time application may increase the processor temperatures to levels at which the reliability of the underlying semiconductor devices is in jeopardy. In this work, I first present a model- based analysis of the thermal impact of a selected periodic task schedule. The proposed thermal analysis method is then extended to estimate the effect of executing a given aperiodic task at a specified time interval along with the periodic task schedule. Thermal sensors are periodically read to correct temperature estimation errors and account for process variations. One of the key features of the proposed method is that it can accurately account for the variations in power consumption during the execution of a task instance. The proposed method can be used at runtime to efficiently schedule an aperiodic task instance without violating its deadline constraint and the temperature constraints of the processor. The effectiveness of the proposed methods is compared with other approaches which do not consider temporal variation of task power consumption